I think what happen in practice is that the decoders still speculatively decode in parallel then drop all misdecoded instructions. Easy when instructions have o

I think what happen in practice is that the decoders still speculatively decode in parallel then drop all misdecoded instructions. Easy when instructions have only a couple of sizes. Hard and wasteful for something like x86.